Texture Contact Lens Detection Based On Deep Learning Technique

Presently biometric recognition patterns have been widely used in the fields of security applications,and iris recognition,as the most securable and stable new recognition pattern among of them,has been strongly concerned,the research on iris recognition has been in a rapid ur development stage.There are still a lot of problems to be solved.Meanwhile,the extraction of iris texture from colored contact lens becomes a serious issue to be dealt with.In this thesis,we focused on extracting iris texture and selecting the colored contact lens image from the real iris under the various complex illuminant.Up to now,most of the pupil detection algorithms rely on the traditional machine learning algorithm for feature multiple extraction and the integration of multiple weak classifiers into strong classifiers.They can not work well due to the detection inaccuracy caused by complex illuminant.Nevertheless,a new algorithm for contact lens detection is proposed in this thesis by combining traditional clustering algorithm with convolution neural network based on deep learning.The related designations are carried out and the results are discussed and performed.Firstly,the basic principles of iris recognition and pupil detection are described briefly,as well as the characteristics and detection standards of contact lens are introduced.The concepts of clustering algorithm and neural network are summarized.And then the colored contact lens database collection is constructed.The difficulties to extract texture features in the database is analyzed,the algorithm design and flow chart are completed.Furtherly,a variety of algorithms are adopted to detect the indoor and outdoor databases respectively.By comparing the accuracy and stability of different algorithms in different databases,the optimized algorithm is obtained,which is currently applied in practical products and verified by national security standards.